Machine Learning: A Game Changer for Additive Manufacturing QA

As I stated in the original version of this article, at 3D Metal Printing Magazine, each of us has a conception of what the term “machine learning” means. Basically, machine learning is based on the idea that systems can learn from data, identify patterns, and make decisions with minimal human intervention. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves.

Machine Learning is just one utilization of an AI built around the concept of providing machines with plenty of real-time diagnostic intelligence, enabling them to autodidact. AI does play an important role by translating the graphics of a designed part to the language used by the machine to actually produce the parts. Machine learning in any environment, including additive manufacturing (AM)/3D metal printing, depends on quality data and the right system to validate that data. The expression “garbage in garbage out” (GIGO) definitely applies.

Is machine learning working in the real-world of additive manufacturing/3D metal printing? The answer is a resounding yes. In a recent article, 3DPrinting.com reported, “Porosity and other defects are a problem where it comes to parts printed with metal powder bed fusion processes. One team of researchers from the U.S. Department of Energy’s Argonne National Laboratory and Texas A&M University has published their findings that may play a part in reducing these subsurface defects. The team has figured out a novel way of predicting the formation of sub-surface porosity as well as measuring the temperature of a region at the very moment of printing of Ti-6Al-4V powder.

Think of the end-to-end part/process quality decision methodology as an ecosystem that functions best if you can distinguish between part and process quality. Effective AM machine learning models are designed to both recognize anomalies in specific parts and monitor the quality of the process itself.

Maintaining the process of examining parts, training models, and monitoring new builds forms a cyclic quality decision ecosystem that contains three components. The training process selects model types from the warehouse, trains them to recognize data patterns in the training builds, and saves the trained models back to the warehouse. The quality assurance process then selects trained models appropriate to material and design of testing builds in order to evaluate ongoing process quality. When new designs and materials are used, the process cycles back to improve the ecosystem.

Following is a diagram showing the part/process quality ecosystem from Sigma Labs.

The PrintRite3D system monitors the builds that are going to be used in training. It produces in-process quality metrics (IPQMs) that will be used as features of machine learning models. Parts of the builds are selected for non-destructive testing (NDT). This testing will be in the form of post process computed tomography (CT) where part porosity is digitally captured. Porosity geometric location data from this analysis is registered and aligned with the Sigma metrics. The tuning and training process uses the metrics as features in machine learning models, and it uses the anomaly data from the process analysis as labels for supervised learning.

The training process is monitored by assessing the diagnostic accuracy with visualization and numerical scores. This information is utilized to further tune the models and their training parameters and complete the description of the training process starting with the monitoring of builds and the detection of anomalies by physical analysis of the parts.

The system now moves from the training process to the quality assurance process. The PrintRite3D system monitors production builds and produces in-process quality metrics (IPQMs) that will be used as features of the trained models. The machine learning quality assurance process feeds the Sigma metrics to trained models in order to evaluate the build quality. Visualization of the predictions is used to identify the location of potential anomalies that are predictive of real defects.

Several tools in the part quality decision dashboard help evaluate the severity of the potential anomalies while the 3D visualization reviews overall patterns, and the SPC chart provides data trend analysis. This completes the description of the quality assurance process starting with the monitoring of builds, the prediction of potential anomaly locations, and combining this information to evaluate the severity of the potential anomalies.

The quality assurance process uses trained models selected from the model warehouse and Sigma production metrics to predict potential anomalies and you are now able to see how the components of the part/process quality decision ecosystem work together. When new designs and materials are used, the process cycles back to continually improve the ecosystem.

Because of the benefits in financial and time savings as well as future cost-effectiveness, machine learning will have a continuing positive impact both for individual manufacturers and AM/3D metal equipment providers, and will also benefit the entire additive manufacturing industry.

Note: the original (and longer) version of this article appeared in the Fall 2020 edition of 3D Metal Printing Magazine.